Addition of cheese whey to raw milk is an extension fraud, and this paper develops a rapid method for its detection. FTIR (Fourier-transform infrared) spectroscopy of milk produces a large amount of data, which can treated by machine learning methods such as classification tree and multilayer perceptron neural networks (MPNN) the two methods used in this study. A total of 520 samples of milk adulterated with cheese whey in concenrations from 1-30% were prepared, and 65 samples were taken as the control. These were stored at different times and temperatures, and analysed by FTIR. A further 520 samples of authentic raw milk were used, and selected components (fat, protein, casein, lactose, total solids, and solids nonfat) and freezing point (°C) were predicted using FTIR, then used as input features for the machine learning algorithms. Performance metrics included accuracy as high as 96.2% for CART (classification and regression trees) and 97.8% for multilayer perceptron neural networks, with precision, sensitivity, and specificity above 95% for both methods. The authors make a caveat on these results that the samples were all prepared from bulk raw milk, not individual milk, whose composition is more variable.
Read the full open access paper
